Skip to content
ANIRUDDH ATREY
QR Code - Contact Aniruddh Atrey
Founder, Product Lead & Full-Stack Engineer · 2026

OutBrew

MetaMinds — Flagship Product

1 0 0 0 0 1 1 1 1 0 0 0 0 1 1 1 1 0 0 1 1 1 1 0 0 1 0 1 1 0 1 0 0 1 1 1 1 1 0 1 0 0 1 1 1 0 1 0 1 1 1 0 0 1 0 0 1 1 1 1 0 0 1
01

The Challenge

The cold email software market is a $1.5B industry growing to $3.2B by 2033 (9.4% CAGR), yet the highest deliverability feature score across all existing competitors is only 38% (8/21 features). Incumbents like Instantly ($20M ARR), Smartlead ($14M ARR), and Lemlist ($40M ARR) optimize for sending volume but fail at the hard parts: inbox placement, personalization at scale, AI-driven research, and follow-through. Sales teams spend more time fighting spam filters than talking to prospects. The opportunity: build a platform that treats deliverability, personalization, and warmup as engineering problems — not growth hacks.

02

The Approach

Architected OutBrew as an enterprise-grade outbound engine across two massive codebases — a Next.js 14 frontend (96K+ lines of TypeScript, 37 dashboard pages, 145 React components) and a FastAPI backend (138K+ lines of Python, 53 API endpoints, 85 backend services, 38 database models). Built on a 9-layer extraction pipeline with 28+ ML/DL services (BERT NER, FAISS, MinHash LSH, Computer Vision), a god-tier caching system with 7-day TTL and freshness indicators, document management with AI parsing, and SSE-streamed real-time progress. Every feature competes head-to-head with Instantly, Smartlead, and Lemlist — and stacks them in a single dashboard.

Architecture

System Architecture

Three integrated layers working as one platform — each designed to be production-grade from day one.

Frontend

Next.js 14 Dashboard

37-page dashboard with 145 React components built on Next.js 14 App Router, TypeScript, Tailwind CSS, Framer Motion spring physics (stiffness 400, damping 15), Radix UI primitives, Zustand state management, and React Hook Form + Zod validation. 96,071 lines of TypeScript.

Next.js 14TypeScriptTailwind CSSFramer MotionZustandRadix UIAxios
Backend

FastAPI + PostgreSQL

FastAPI 0.104+ on Python 3.11+ with 53 endpoint files, 85 service modules, and 38 SQLAlchemy async models. PostgreSQL 15+ with distributed Redis rate limiting, JWT auth with refresh tokens and blacklisting, and SMTP sending with retry, open tracking, and click tracking. 138,049 lines of Python.

FastAPIPython 3.11PostgreSQL 15SQLAlchemy 2Redis 7JWTSMTP
AI Layer

Extraction + Research Engine

9-layer extraction pipeline covering 4 sectors (Clients, Companies, Recruiters, Customers) with 28+ ML/DL services including BERT NER, FAISS vector indexing, MinHash LSH deduplication, and Computer Vision. Anthropic Claude + OpenAI GPT integration for research and email generation. Playwright for company scraping.

Anthropic ClaudeOpenAI GPTBERT NERFAISSMinHash LSHPlaywrightSSE
03

The Solution

  • Built a full authentication stack with JWT + refresh tokens, email verification, rate limiting, and token blacklisting
  • Engineered a 9-layer extraction engine handling URLs, files, CSVs, and API integrations (Apollo.io, Hunter.io) with real-time SSE streaming
  • Designed a 6-step extraction wizard: Sector → Source → Filter → Progress → Review → Integration
  • Built a god-tier cache system with 7-day TTL, color-coded freshness indicators (Green/Yellow/Orange/Red), and full version history
  • Developed a document management system with AI-powered resume and info document parsing, PDF/image preview, and bulk operations
  • Integrated follow-up queue into Recipients page with auto follow-up settings, bulk send, and template variables
  • Implemented email warming with strategy presets, template marketplace with conversion scoring, and A/B testing framework
  • Created company intelligence module with Playwright-based scraping and AI-powered research summaries
  • Added rate limiting with daily/monthly enforcement, warmup health monitoring, and send-time optimization
  • Built 47 sample applications for intelligent dashboard fallback — solving the 'empty state' problem elegantly
Features

Feature Breakdown

Extraction Engine

  • 4 target sectors: Clients, Companies, Recruiters, Customers
  • 9-layer pipeline from static HTML to ML-powered entity resolution
  • SSE streaming with <1s latency real-time progress
  • 3-sheet Excel export with quality reports
  • One-click integration to Recipients table
  • Apollo.io and Hunter.io API integrations

AI Research & Email Gen

  • Cached company research with 7-day TTL
  • Freshness indicators: Green → Yellow → Orange → Red by age
  • Email library with search, filter, and favorites
  • Max 10 cached batches per recipient
  • Regenerate button with confirmation
  • Version tracking for all research and emails

Campaign & Follow-up

  • Follow-up queue with 3-30 day configurable threshold
  • Auto follow-up settings with scheduling
  • Bulk send with template variables ({name}, {company}, {days_waiting})
  • Email warming with 3 strategy presets
  • Template marketplace with conversion scoring
  • A/B testing framework with statistical analysis

Document Management

  • AI-powered resume parsing (skills, experience, achievements)
  • Info document parsing for company context
  • PDF preview with fullscreen mode
  • Image preview (jpg, png, gif, webp)
  • Bulk operations: delete, re-parse, export
  • Context selection for AI personalization
Scale

Scale & Scope

96K+
Lines TypeScript
138K+
Lines Python
37
Dashboard Pages
145
React Components
53
API Endpoints
85
Backend Services
38
DB Models
28+
ML/DL Services
04

The Impact

0 Layer Extraction
0 Target Sectors
0-day Cache TTL
0 Current Release

Tech Stack

Next.js 14 TypeScript 5 Tailwind CSS 3 Framer Motion 10 Zustand 4 Radix UI React Hook Form Zod Axios FastAPI 0.104 Python 3.11 SQLAlchemy 2.0 PostgreSQL 15 Redis 7 JWT Anthropic Claude OpenAI GPT Playwright BERT NER FAISS SMTP SSE
“Outbound is not about sending more email — it is about sending email that arrives, gets opened, and actually converts. OutBrew is what happens when warmup, personalization, research, and follow-up are treated as engineering problems, not growth hacks. Two hundred thousand lines of code, 145 components, 53 endpoints, and a 9-layer AI extraction engine — all designed to take on a $1.5B market dominated by platforms that stopped innovating at the template level.”
Next Case Study VelaReach →